Modelling of Normal Tissue Complication Probabilities (NTCP): Review of Application of Machine Learning in Predicting NTCP

Abstract

Predicting normal tissue toxicity following radiotherapy is a multidimensional challenge. The dose received by healthy tissue surrounding the tumour is described using a 3D dose distribution. In addition, patient-and treatment-related factors must also be considered in any predictive model of toxicity. Mixing these complex and disparate data types is a challenge that can be addressed with machine learning. This chapter introduces the concept of normal tissue complication probability (NTCP) and reviews literature related to the use of machine learning in this field. NTCP ModellingThe response of normal tissue incidentally and unavoidably irradiated during radiotherapy is the main factor limiting the increase in prescription dose to the tumour. Optimising this trade-off, known as the therapeutic ratio, is the fundamental challenge in radiotherapy (Fig. 17.1). Although complimentary in approach, the complexity of predicting normal tissue response is a higher dimensional problem than predicting local control. The reasons for this are (1) there are usually more than one organ at risk irradiated and protecting all of these structures requires compromise, (2) each structure responds differently to radiotherapy due to the type of cells and the structural and functional organisation of the tissue, and (3) the dose distributions to the surrounding normal tissues are inhomogeneous with gradients

Using machine learning algorithm to analyse the hypothyroidism complications caused by radiotherapy in patients with head and neck cancer

“…The Lyman Kutcher-Burman (LKB) 8 algorithm is often used clinically to assess the normal tissue complication probability (NTCP). However, the standard deviation of the data samples in this study was too large to use the NTCP algorithm in an optimal fashion 9 . Supervised machine learning was chosen for use instead.…”

Modelling for Radiation Treatment Outcome

“…For this purpose, the three-dimensional dose distribution is often reduced to a few simple metrics that can be derived from a dose-volume histogram (DVH). Some of the different methods for modelling clinical outcome data of retrospective patient cohorts and their dose distributions are described as follows [78]. The parameter n describes the volume effect of the investigated OAR [84].…”

“…Tissue-Architecture Models: These more mechanistic models are based on the functional architecture of the tissue by introducing functional subunits of an OAR. These can be anatomical substructures, such as nephrons of the kidney, or the largest cell group that still functions as long as it comprises a surviving clonogen [78]. These functional subunits can be arranged in serial or parallel order, or in a combination of both.…”

Using machine learning algorithm to analyse the hypothyroidism complications caused by radiotherapy in patients with head and neck cancer

“…The Lyman Kutcher-Burman (LKB) [3] algorithm is often used clinically to assess the normal tissue complication probability (NTCP). However, the standard deviation of the data samples in this study was too large to use the NTCP algorithm in an optimal fashion [4]. Supervised machine learning was chosen for use instead.…”